Preprocess Features for Machine Learning

This function preprocesses a dataset by removing features with near-zero variance, highly correlated features, and features that are constant within any class of the target variable. It ensures that the resulting feature set is more suitable for machine learning models.

Usage

preprocess_features(data, target_col = "target", cor_thresh = 0.9)

Arguments

data

A data frame containing predictor features and the target variable.

target_col

A character string specifying the name of the target column in data. Default is "target".

cor_thresh

A numeric value between 0 and 1 specifying the correlation threshold for removing highly correlated features. Default is 0.9.

Value

A data frame with the preprocessed features and the target column.

Details

The preprocessing steps include:

1. Removing near-zero variance features (using caret::nearZeroVar).

2. Removing highly correlated features above the specified threshold (using caret::findCorrelation).

3. Removing features that are constant within any class of the target variable (i.e., provide no discriminatory power across classes).

After preprocessing, the target column is re-attached to the dataset.

Examples

if (FALSE) { # \dontrun{
library(caret)
library(dplyr)

set.seed(123)
df <- data.frame(
  feature1 = c(1, 1, 1, 1, 1),             # constant
  feature2 = c(1, 2, 3, 4, 5),             # numeric
  feature3 = c(1, 2, 3, 4, 5) * 2,         # highly correlated with feature2
  target   = c("A", "A", "B", "B", "B")
)

clean_df <- preprocess_features(df, target_col = "target", cor_thresh = 0.9)
} # }